home page about us documents miscellaneous sitemap
home page about us documents miscellaneous sitemap

Mission Impossible? User Meets Agent
  Hyacinth S. Nwana & Divine T. Ndumu (Hg.), Proceedings of the Third International Conference on the Practical Aplication of Intelligent Agents and Multi-Agent Technology, The Practical Application Company Ltd., Blackpool, S. 173-189, 1998.

Ute Hoffmann, Jillian-Beth Stamos-Kaschke , 4/98

  jump-off point
1  1. Introduction
2  2. Shaping artifacts and users
3  3. Envisioning users and agents
4  4. Using agents
5  4.1. Firefly/MyLaunch
6  4.2. Travels with Charley - A traffic agent
7  4. 3. Non Web-based agents
8  5. Conclusion


1 1. Introduction
  As predicted in the early '90s, "Agent-based computing (ABC) is likely to be the next significant breakthrough in software development." (Sargent 1992) A growing number of agent-related conferences and events (http://www.agent.org/pub/events.ht ml), a range of recent publications (http://botspot.com/bookstore/in dex.html) and a proliferating agent industry (Guilfoyle et al. 1997) are clear signs that agent technology and its applications are flourishing.

Yet in the November 1996 issue of Wired, web agents figured on the hype list with an expected lifetime of four months. Steve G. Steinberg, the author of the hype list, uses pleasure of process as an argument that those who advocate intelligent agents rarely understand. According to Steinberg, web agents, which troll the Web looking for items that match your stated interests, suffer from an obvious problem: few things on the Web are really interesting. It's the process of looking -- of following links and being surprised by strange juxtapositions - that is enjoyable.

Whether the process of looking for things is really as enjoyable to Web users remains to be seen. The fact is that as well as Web agents, there are a number of other agent types and uses. We quoted Steinberg's argument so extensively because he draws the attention to on an important but elusive figure in agent-based computing: the user.

In this paper, we focus on the user from different perspectives. In the following part of the paper we very briefly outline a conceptual model of how not only objects but also (projected) users are shaped in the process of designing technological artifacts. We then deal with a popular form of making a new technology attractive to (real) users: scenarisation. In the next part a model user is heard and reports on some of her experiences made using the following agents: Firefly (MyLaunch), TravAgent and Microsoft Bob. Obviously, we did not choose these agents because they represent the cutting edge of agent technology. We chose to deal with them because these agents are accessible to a wide range of users and existing applications such as these help form the image of agent technology outside the agent technology community.

2 2. Shaping artifacts and users
  Very few designers of agent systems would deny the importance of users and that they need to be taken into consideration when building systems. Designers seek - or so they say - to address the perceived needs of users who are taken for granted. In this view, there is a more or less fixed world out there to be represented in the design process. The problem then is of how to find adequate or "faithful" representations.

Contrary to this representation model of the end user we suggest an approach that can help describe the shaping of the user in the design process. A large part of the work of designers is that of inscribing their vision of (or prediction about) the user - as well as other entities that make up the world into which the system or programme is to be launched - in the technical content of a software tool.1 Ideas turn into objects through a series of negotiations in which actors of all kinds play out their lives. The proposed associations between the different entities concerned (such as electrons, communication protocols, industrial firms and consumers) are heterogeneous from the start of the process. To describe these heterogeneous associations, the notion of an actor network was introduced.2

A technological project is negotiated and transformed as it moves "downstream" under the influence it seeks, although not always successfully, to enrol. The proposed associations, and, by consequence, the project itself, will hold together only if the different entities concerned can be persuaded to accept the roles assigned to them. The initial gadget moves only if it can interest one group or the other to play the roles proposed for them. It may be that no actors will come forward to play the roles envisaged by the engineer or designer. At the end of the road, users may define quite different roles of their own. If this happens, the object remains a chimera, for it is in the confrontation between technical objects and their users that the latter are rendered real or unreal. The insistence on process in framing emerging technologies means, for instance, that no agent is ever complete, autonomous, and final. Another way of saying this is to note that the bits and pieces assembled into an order are constantly liable to break down, or make off on their own.

In view of this approach to appreciate what it might take to realise a successful application of agent technology the question is not whether designers can create intelligent agents along preconceived technical notions but rather whether such agents eventually face the sort of world required for them to be effective. In other words: "Rather than simply developing agents and setting them in place in the world, the world and the people in it have to change in order for the agents to really do want we [the designers , J. S.-K., U.H.] want them to do." (Horberg 1996: 73)

3 3. Envisioning users and agents
  So far we have argued that the realisation of a technological project depends on the association of (potentially unhelpful) elements into self-sustaining networks that are able to resist dissociation. This includes the work of contextualisation. A technological project is not taking place in a context: it gives itself a context. More often that not, the first result of contextualisation is a preliminary scenarisation. Scenarios or "scripts" provide narrative programs of action that mirror the designer's beliefs about the relationships between a technical artefact and its surrounding actors. Scenarisation is thus an attempt to predetermine the settings that users are asked to imagine for a particular piece of technology and the prescriptions that accompany it. 3 Scenarios blend together major social questions concerning the spirit of the age and "proper" technological questions in a single discourse. In the field of agent technology, scenarios are also a good source for studying the designers' projected user.4 A developer's original visions might not be accessible as they were not published. In this case, the media could act as our source for this paper. It often happens that journalists are the mouthpiece of innovators or are treated as such.

By chance, we happen to have an article at hand from the French magazine l'Express dated June 5th 198 1, in which two journalists weave their impressions taken from a journey to the USA into the following scenario:

The electronic home of Mr. and Mrs. Smith 5

"Imagine the following situation: At exactly 7:30, the bedroom curtains are drawn, the needle on the thermometer goes up to 28 'C, the coffee machine in the kitchen starts hissing, the door leading to the garden opens to let the dog out and the television starts broadcasting the first news of the day. We are at the home of Mr. and Mrs. Smith, a middle-class couple. On the screen of their home computer, Mrs. Smith compares several local supermarkets' prices of things she has to buy. At the touch of a button, she calls up the recipe for boeuf bourgignon on the terminal in the kitchen, telling the machine to work out the ingredients needed for six people. She programmes her stove so that each hob will have the right temperature at 7.15 p. m. Then, she takes a shower which has been set to the right temperature the night before. Afterwards, she sits down in front of the television to take part in a discussion about Byzantine art, in the subject of which she has been instructed by her computer. She gives her daughter, who has been praised by the computer for doing her homework well, a hug. Then she goes out, to return only minutes before her guests arrive.........."

In all probability you will not have any difficulty in recognising this description as being a typical example of a scenario. You will have noticed that he scenario does not rely on agent technology at all. This could be due to the fact that in the early Eighties, agent technology had not yet made a strong appearance. Another important reason is that scenarios as programmes of action tend to be not too clear about the precise technologies involved in the envisioned actor world. This is advantageous, as in the beginning it is appropriate for different groups with divergent interests to embark on a project that constitutes a good trading-post for goals with a certain amount of vagueness. Narrative programs of action, and the technological projects they are related to, become irreversible in the same measure as words are inscribed into another matter, as texts are transformed into machines.6 Emerging technologies can be inscribed into the same scenario over and over again. Let us assume that the authors of the scenario described above had gone on the same journey, but 15 years later, i.e. at some point during the middle of the Nineties. We have taken the liberty to assume what form their report would have taken, accounting for agent technology hype:

The wired home of Mr. and Mrs. Smith:7

Imagine the following situation: At 8:30 a. m. precisely, the blinds in the bedroom are drawn; sensors calculate the amount of pollution in the flat,. the intelligent central heating readjusts itself according to outside temperatures and minimal fuel consumption: the coffee machine hooked up to the Internet starts hissing, the Tamagotchi starts peeping and needs feeding, and the first personalised news of the day starts broadcasting on the Net Computer via a push canal We are at the home of Mr. and Mrs. Smith, a middle-class couple. On the screen of her multimedia computer, Mrs. Smith has [agent] compare the prices of several local supermarkets for things she has to buy. At the kitchen terminal she is shown a menu for the dinner party she wants to give that evening by [agent], taking her guests' favourite food and what vegetables are in season into account. Using speech recognition, she tells the machine to calculate the amount needed for eight people and programmes her microwave to turn itself on at the right temperature at 7:15 p. m. She then takes a shower which has been set to the right temperature the night before. Afterwards, she reads her email, which has been automatically sorted by [agent] and takes a look at the WWW housewife club pages to see if there is anything new in the frustrated wives' forum which she moderates. She hugs her daughter, who has just been praised by [Microsoft Bob] for doing her homework well. After asking [TravelAgent] for an outlook on afternoon traffic jams and an optimal route for her journey home, she goes out, only to return to her smart home minutes before her guests arrive.

Much has become different in this second scenario, but hardly anything has really changed. One thing seems to be certain: Nothing is possible without intelligent agents - or is it?

Now at the latest, one may be tempted to ask of what the notion of agent "really" means. At its current state of development, however, agent technology has neither an accepted definition nor a clear set of design principles for building software or hardware agents: "There is no general agreement as to what constitutes an agent, or as to how agents differ from programs." (Franklin & Graesser 1996). For a field of technology which is still at a nascent stage, this may not seem at all surprising. In part, this lack of terminological clarity can even be regarded as conducive due to the large but unclear number of agents to be mobilised. In the beginning, it is appropriate for different groups with divergent interests to conspire with a certain amount of vagueness on a project that constitutes a good trading-post for goals.8 Even though in the agent-based community the question as to what an agent is, technically speaking, is controversial, one of the prevailing assumptions is that agents are useful. Some examples of typical statements directed towards users can be found in agent literature, such as: "Intelligent Agents are software entities that assist people and act on their behalf. They make your life easier, save you time, and simplify your complex world." Or: "Think of an intelligent agent as a robot -- a software robot. Give it instructions, set it loose on the Internet, and your intelligent agent will gather up for you information about jobs, bargains, hobbies, companies whose stock you own -- in short, any information you want." Whether or not applications of agent technology do have the potential to become a completely natural part of everyday life is a far from trivial question. What makes agent technology so seductive to the user? Wherein lies its irresistible promise?

Everyday tasks are usually routine ones, which do not continually redesign actions as a number of steps which have to be taken anew each time a new task is commenced, but call upon typical actions or types of action sequences habitually (Hennen 1992). Agent technology presents itself to potential users as a predefined method of problem-solving. Agents appeal to a user's predilection for activating typical patterns of behaviour under typical circumstances without prior planning. Agent technology is not suited to open situations in which the actor can choose another objective or another action plan.

Uses of agent technology are designed to solve a problem in a specific situation. The software agent contains the correlation between problem and solution, which is objectified in a certain manner. Software agents form an objectified action plan. From the user's point of view, agents are a problem-solving suggestion which has to be acquired in an interpretative manner and possibly modified.

In order to be successful, agent technology has to seduce and recruit the user to raise her interest. User demand and user interest are negotiable like everything else and shaping them constitutes an integral part of projects in the field of agent technology. As a rule, the user is modelled as being enthusiastic. But will flesh-and-blood users subscribe to the designers' version of them, and settle nonchalantly into the comfortable spot that experts have spent months or years preparing for them? Let us follow some of the actors and look at their reaction to what is prescribed to them. In doing this, we leave the world inscribed in the object and enter the world described by its displacement.

4 4. Using agents
  One of the principal uses for agents is as an information manager to filter and obtain information on their users' behalf. MAXIMS, an email agent, was developed by Pattie Maes (1994) [http://pattie.www.media.mit.edu/people/pattie/CACM-94/CACM-94-pl.html] at MIT which predicted what the user would do with incoming email and made suggestions accordingly. The more familiar it became with its user's actions, the easier it was for it to anticipate its next step, thus reducing the user's workload.

Due to the popularity of the Internet and especially the World Wide Web, it has become increasingly difficult to keep up with changes and find relevant information. Ideally, a personal Web agent would know how to sort through the plethora of information it came up with instead of leaving the usually laborious and time-consuming task of asking various search engines to the user, as the hits offered by one search engine may not be comprehensive and it is usually necessary to consult at least two search engines to obtain useful results. A growing number of companies has sprung up, offering agents which ensure every user receives information that is tailored to her needs and can be easily located. Not only that, but agents save time, something most businesses are interested in, as well as people looking for information from home - not only time, but money and human resources. While an agent is scouring the Web for information, the user can get on with less menial tasks until it returns.

Now agents have progressed from being a wistful gleam in a programmer's eye to something touted almost everywhere - there are shopping agents, travel agents, agents that provide people with musical suggestions, all out there just waiting to facilitate one's journeys through the Web, feeding one with choice morsels of information, why isn't everyone using them? Why do people still primarily rely on search engines when they are looking for something and sort through their mail themselves?

After having read and heard so much about agents, we decided to put them to the test. When testing software, there are several possibilities to choose from. The first and probably one of the most popular is the questionnaire. A large target group, ideally at a similar level of competence, is asked to use a product and afterwards, answer questions. This method is most suited to questions that can be answered simply by checking a box or giving a brief answer. This method is recommended if the usage of the product lends itself to being documented using the question and answer method. We did not necessarily want to know, for example, whether it was easy to use an agent, but rather the results obtained by using it. A further problem is the fact that agents are not yet widely-used, which would mean an accordingly small group of representative users from whom to obtain relevant data.

Another popular method for testing software is asking users to interact with it in a usability lab. This is especially helpful when testing software before putting it on the market. Users are watched through the entire process of interaction through a two-way mirror or pane of glass. We felt the problem with usability labs was that we wanted to test agent software that was already being used rather than at an alpha or beta stage. How users interacted with agents was secondary in this case; our emphasis was on whether users could benefit from working with agents and not how they went about it.

Our method was to choose a computer-literate colleague with a profound knowledge of the Web and computers in general and ask her to test three of the most popular Web agents as to whether or not they fulfil users' expectations. We explicitly chose this method as we believe it to be superior to the others in this case, as we wished to find out more about exactly where problems could arise. This, we felt, could be achieved most effectively by having the user document the entire process herself. Here is how she fared with some of the most popular Web agents.

5 4.1. Firefly/MyLaunch
  I tried out some of the more well-known agents to see what they did and how they did it. After all, having a personal agent to assist me on my merry way sounded very flattering after the anonymity on the Web I'd hitherto been used to. First, I paid a visit to Firefly which is aptly named, as it regularly undergoes a complete metamorphosis. Formerly known as HOMR and RINGO, it was developed at MIT and is based on the technique of collaborative filtering. Firefly retrieves its information from a relatively limited database consisting of descriptions of artists and their musical offerings rather than the unlimited offerings one usually encounters on the Web, which should ensure that information is processed more quickly and efficiently. Irrelevant information which always forms part of a Web search is not to be found here. On the other hand, as the database is limited, this means it is also incomplete to a certain extent.

Firefly recommendations are taken from correlations between users; i.e. if Jane Musiclover enters the Rolling Stones as being her favourite group, she is presented with 5 groups to rate, using a scale of 1 (hate it!) to 7 (the best!) which other Rolling Stones fans using Firefly also like and she, being a Stones fan, should too. Her ratings are recorded and stored for later use when she next visits the Firefly site. Eventually, by learning from her tastes and comparing them to other people's, Firefly "knows" what other groups she could like. The more often she visits the site and rates bands, the easier it is to fine-tune her profile. To facilitate the process, sound samples can be listened to and there are album and artist ratings by other Firefly members which can be read. Bands can also be rated directly rather than having Firefly suggest them, so if Jane absolutely hates Michael Jackson, she can go directly to his profile and give him a 1. In a similar vein, she is also able to go directly to the Rolling Stones profile and not only give them a 7, but add them to her list of favourites. If she should so wish, she can review and rate any of their albums or read other people's reviews and ratings. This provides for a more specialised profile, as not only individual groups are rated, but their musical offerings as well.

As well as looking for music, Jane can look for Firefly members that share her musical tastes and go to their profiles. Firefly informs her as to whether they are currently online (i.e. accessing Firefly) and if she should so wish, she can chat to them, leave them a message or add them to her list of friends. This is something a search engine could never do: "Please find bands that are similar to Jimi Hendrix and people who like similar music to myself'.

http://firefly.com redirected me to http://firefly.net (see what I mean?) which looked nothing like the site I'd been used to. Web mutatur et nos mutamur in illa. I entered my Firefly alias and password and was informed that what used to be known as Firefly was now called myLaunch, which could be found at http://www.mylaunch.com and for some reason refused to recognise me. I therefore created another alias and started all over again. The page told me to be patient, as it was "creating my customised myLaunch page." After waiting patiently I was rewarded with a page called MyMenu. This contained music news, tour dates and new releases and a "customize" button. Obviously, in order to be able to provide me with a page tailored to my needs, it needed to know a little more. Unfortunately, all that was meant by customising related to exactly how many articles I wished to see when accessing the MyMusic page and also whether I wished to access live radio stations (see Figure 1). What I'd expected was something more along the lines of whether I wanted news of new releases, certain bands, tour dates and music news or even if I would like categories to choose from, which news services such as CRAYON offer. This wasn't based on the principle behind CRAYON, however, and I wasn't really looking for music news anyway, I wanted further information to whet my appetite for bands I hadn't heard of or had heard of but didn't know anything about etc. Admittedly, my taste in music veers towards the eclectic at times, but that was where services like MyLaunch came in, or so I thought.

Figure 1
MyMusic customisation

After my first disappointment, I decided to ask MyLaunch to recommend artists, which it duly proceeded to do. After all, that's how it learns about what I like. After about thirty minutes of rating albums and artists, I was rewarded with the message that MyLaunch didn't have enough information about me and would therefore like me to "rate a selection of popular music" which I did willingly. The same screen, complete with my ratings, appeared again with the same message. I then looked for artists whose music I liked in the hope that MyLaunch would finally know more about me and we could carry on. Unfortunately, MyLaunch still wanted more information, and ultimately ran out of suggestions. As I had spent over two hours rating and re-rating, I gave up.

Figure 2
MyLaunch needs more information

MyLaunch is obviously geared towards people who aren't looking for anything in particular and have enough time and patience for it to learn about a user's likes and dislikes. Rather than rely on the data a serendipity search with a search engine will come up with, users can go straight to MyLaunch and receive detailed information about artists which will be recorded and added to their profile. An interesting aspect was that MyLaunch suggested artists I would not have expected; after giving Frank Zappa a 7, for example, I was given 5 jazz artists to rate, rather than someone like Captain Beefheart who springs to mind more readily.

The question that arises after using MyLaunch is: can this be called an agent? Its approach may be more personal than a search engine's, but it needs so much input from the user that it is unlikely to be widely used. Even after informing it of one's preferences, obtaining recommendations remains difficult. Does it really learn? Is this ability a useful feature? Or is MyLaunch just used as an AOL user profile might be used: as a possibility of informing others as to one's likes?

What would a more static medium such as a CD-ROM have had to offer? Microsoft's Music Central '96, when asked about artists most like Frank Zappa, immediately came up with Captain Beefheart, closely followed by the Residents and John Cage (see Figure 3)

Figure 3
Music Central screenshot

Music Central only offers this possibility for a limited number of artists who are either representative of a particular genre or particularly famous. Each user can make lists of her favourite artists, although a rating system as can be found at MyLaunch is not possible. Artist and album reviews are not dynamic, but come with the CD-ROM, although it is possible to download updates once a month from Microsoft's website, thus leaving the decision in the hands of the user. As does MyLaunch, Music Central culls its information from a database which is subject to the same limitations as regards completeness.

Admittedly, Music Central is a commercial product and thus more costly than placing a call to one's Internet provider, but saves time and produces visible and quick results.

6 4.2. Travels with Charley - A traffic agent
  One of the most popular uses for agents seems to be as traffic wardens - you tell your agent where you're going and, ideally, it will inform you as to the conditions you should expect on the way. For instance, when driving from Seattle to Portland, it would be nice to know about road conditions in advance - whether to expect snow, rain or fog, what alternative route to take in case of traffic jams or whether it might even be better to go by train. An ideal personal traffic agent would "know" without any further prompting from you which routes you habitually take and, by interacting on its own behalf with other agents, be able to inform you as to whether your favourite shortcut is chockablock with traffic (and therefore not much of a shortcut) and, if so, suggest alternatives. Should you find yourself in a strange town, your agent will guide you safely to your destination without your ever having to take one look at a map, thereby considerably lowering the chances of an accident. Before meeting a friend at the airport, it would be helpful to know if the flight is on time, if the roads leading to the airport are free, if public transport is running to schedule etc.

Travel Agent is a Java-based agent that currently only offers traffic information for the US. The user can choose between interactive and noninteractive TravAgents, the difference being that non-interactive TravAgents only list predefined services as offered by various public transport authorities in a certain area (for example, all the bus timetables for New Jersey) and interactive TravAgents allow the user to search by state and method of transport used (car, bus/rail, water, air). In addition, it is also possible to specify whether maps, text, video or audio should be used. Multiple entries are possible. There is also an extra menu for such things related to travelling such as car hire, restaurants, hotels, service stations and weather updates. In addition, road construction reports are taken into account, as are traffic cameras.

Once the user has stated her preferences, TravAgent searches the World Wide Web for information. Initially, this is a lengthy process. Once the information has been received, however, TravAgent is surprisingly fast and efficient. In fact, it almost seems too quick to be true. A more detailed look at what goes on behind TravAgent showed that instead of searching the Web every time an enquiry is made, TravAgent draws its information from a static database which is downloaded during the initial search, which explains the delay at the beginning. The database is comprised of URLs pertaining to travel in different American states and cities and, although checking different criteria gives the user the illusion of accessing a dynamic online database, does nothing but include URLs that comply with what the user wishes to know and discard URLs that are irrelevant to the search term in question. Therefore, checking "Alaska" and car', "bus water", "air" and all the possible media types will result in TravAgent giving all the information it has in its database about travel in Alaska. Removing categories will not mean a more thorough search in the remaining fields, but rather that TravAgent simply removes URLs from its list.

As TravAgent does not provide information itself but merely links to other sites instead, there is no way of knowing how up to date the information is. Users have no idea how old the data is that they are accessing and there is nothing that TravAgent has to offer that a search engine could not come up with as well.

Additional information such as restaurants, car rental firms etc. is only given if the business in question has a home page on the WWW and directs the user to the main page rather than giving regional information.

If anything, TravAgent is a specialised search engine for people interested in travelling through America who do not require detailed or precise information. As stated on the TravAgent web page, the idea is that firms or individuals wishing to provide visitors to their pages with traffic information should add a link to the TravAgent applet. The question foremost in my mind was "Can this be called an agent?" Neither is TravAgent particularly interactive, nor is it personal. It has no way of recording a user's preferences and offers nothing that cannot be found elsewhere, although it must be said that the probability of finding such specialised information on a single page is slight. As it is Java-based, searches are carried out from the web page containing the link, rather than on the TravAgent homepage, thus giving the user the illusion that the information he or she is being given is unique to the respective site. It uses a similar approach to Yahoo! in that it categorizes information, which can be helpful under some circumstances. Therefore, along these lines, one could argue that Yahoo! is also an agent.

7 4. 3. Non Web-based agents
  Agents can also be used to interact with inexperienced users at desktop level. As soon as the agent realises a user is stuck or about to do something new, it can pop up and offer help. Windows wizards are an example of this type of agency, as they go beyond mere help screens. Instead of the user explicitly asking for help, the wizards make suggestions themselves whenever it seems appropriate, which is like having someone more experienced at hand, thus rendering the computer more human and less terrifying. Techspeak is reduced to a minimum; things being explained in more easily understood terms.

Microsoft Bob, designed to make PCs more accessible to computerphobes and children and described by Microsoft as a "social interface", was one such example, released shortly before Windows95 and since eclipsed by its success. The desktop was transformed into a "house" in which each user could create "rooms" in which different applications "lived". This caused more serious computer users to scoff Bob off as being more of a toy application than an environment people could actually work with. However, Bob was specifically designed to be a "family PC" everyone could use, even complete computer novices who could carry out tasks without any knowledge of how the computer worked.9

Bob came with various built-in programmes, specifically designed for domestic use, instead of stand-alone applications such as Word or Excel. It was not possible to add multiple columns or footnotes when using the letter-writer, but clip art and borders were available. An email programme was not included but could be purchased separately. Interaction with Bob's interface was facilitated by a guide which could be chosen out of 12 cartoon characters and launched into an uninterruptible cartoon routine, thereby obliging the user to wait until it was finished. Guides could be switched off as soon as a user felt comfortable using Bob without any further assistance. Users could create public and private rooms that could be decorated with items that had no further function, such as pets, wallpaper and furniture. Private rooms could be protected with passwords. When a user created enough rooms, Bob gave them their own house. In addition to the decorations included with Bob, a Plus pack could also be purchased, offering seasonal clip art and more guides.

The design philosophy behind Bob is similar to that of the menus of early word processing programmes such as WangWriter, designed for typists with no previous computer knowledge. Typists could choose between predefined menu items without having to memorise commands unrelated to the work they were doing, the main idea being that operators should have as little control over the system as possible in order to minimise mistakes: "Whatever option the user chooses leads to either prompts or a second menu with further options. In other words, there is not much that an operator can do wrong." (Seybold 1982:3) Users are treated as eternal beginners without there being any possibility of progressing to an advanced level. Users that started out as dummies remained that way due to the static nature of the programme which, together with the increasing popularity of the Internet and especially the WWW, led to its rapid demise.

Instead of an improved and more "intelligent" version being developed, Microsoft subsequently concentrated on making software Internet compatible and on creating more intuitive help wizards, to be used in specific programmes mostly related to the Microsoft Office suite. Tips are given whenever a programme is started (this option can be turned off if the user so wishes) and whenever a user either explicitly requests help or where a shortcut can be suggested. Assistance such as this is preferable to that given by Bob's guides, it being more comprehensive and tailored to a user's individual needs in different situations. Although the interface is less elaborate, it is easier on the computer's resources (when Bob was first launched, one of the major points of criticism was the fact that users were obliged to upgrade their hardware). Dedicated tips are more likely to be accepted by adults who could feel Bob's interface infers that non-computer literate users are not taken as seriously as they should.

8 5. Conclusion
  Agent technology is part of a movement of computers into broader areas of application. How agents might be used to mediate a broader range of interactions and influence relationships is one of the most significant and interesting aspects of the future computing debate. How to produce agents as well as the very notion of an intelligent or autonomous agent has been an ongoing debate ever since the idea was first elaborated upon.

In this paper we have argued that building agent-based systems may fruitfully be seen as a process of interdefinition of human and non-human actors which can be understood both as sociotechnical constructs and must not be treated as given.

Agent production may also be seen as subject to a translation process in which the initially envisioned gadget is negotiated and transformed under the influence of the surrounding actors it seeks to enrol. It is in the confrontation between agents and their users that the latter are rendered real or unreal and agent technology is reinvested and reshaped in use. "User meets agent" also means that eventually the "real" user eventually meets the designers' projected user embedded in the artifact. How this meeting turns out is not least a function of the distribution of competencies assumed when an agent is conceived and designed.

With our paper we set out to draw further attention to some of problem zones regarding the usefulness of agents as seen from the (real) user's point of view.

Software agents form an objectified action plan. Positively speaking, for the user, this action plan can constitute a possibility to act, but can also represent options for acting that are too rigid. If the agent limits a user's possibilities, the user will try to extract herself out of the prescribed program of action or adjust the setting through some negotiations.

This holds even more true if a user does not quite see why an agent could assist her with a task, especially if it is one she can carry out efficiently and quickly herself without first having to go through the laborious process of training an agent. The trade-off between the effort needed and the advantage gained is exacerbated when using commercial agents that cost money. Designers should take the fact into account that users of the possible services an agent has to offer could weigh them up against their own capabilities. The agent paradigm includes the possibility that end users become their own developers to a certain extent, building personalised agent applications. It is unlikely, however, that users would have the patience to go through elaborate agent-training rituals.

The choices made by designers regarding their decisions about what should be delegated to whom or what define the space in which agents and users move. Agents as technical objects contain a "geography of responsibilities" (Akrich 1992, 207) which is open to question and may be resisted.

Agents can be implemented not only to aid users but content providers as well. Drawing conclusions from the way users interact with agents, companies can draw upon this data to create useful "prototype" profiles. These profiles can then be made accessible to other users, thus improving the flow of information. Agents can not only retrieve data, but forward it to other users as well. This possibility could be used to establish virtual communities - agents send data to users interested in a certain subject, they in turn can respond and send more information to others via the agent. However, care must be taken to ensure that agents are not abused as "double agents" - the more information the user gives about herself, the easier it is for other firms to find about her exact online behaviour, consumer habits, in short, all things which can be used for marketing research or for political control as in totalitarian countries. If problems like this occur on a larger scale, it would be a sensible idea to appoint an independent control board for the examination of software agents like consumer boards do with other products or the Food and Drug Administration does with formulas and medical devices and examine products which are already available or which are about to enter the market. The test results could then be distributed throughout the Internet. Agent look-up might be facilitated by agent registries. Another possibility would be a Campaign for Real Agents along the lines of CAMRA. The agent standards authority FIPA might be a first step in this direction, as might the agent language KQML (knowledge query markup language).

  Actor Network Resource. An Annotated Bibliography, 1997. Centre for Social Theory and Technology, Keele University. Version 1. 1.
http://www.keele.ac.uk/depts/s tt/stt/ant/

Akrich, M., 1992a, "The Description of Technical Objects", in: Shaping Technology/Building Society. Studies in Sociotechnical Change, W. E. Bijker & J. Law (eds.). Cambridge, MIT Press, 205-240.

Akrich, M. & Latour, B., 1992. "A Summary of a Convenient Vocabulary for the Semiotics Human and Nonhuman Assemblies,", in: W. E. Bijker & J. Law (eds.), Shaping Technology/Building Society. Studies in Sociotechnical Change. Cambridge, MA London, Uk: MIT Press, 259-263.

Akrich, M., 1992b. "Beyond social construction of technology: The shaping of people things in the innovation process," in: M. Dierkes & U. Hoffmann (eds.). New Technologies at the Outset. Social Forces in the Shaping of Technological Innovations. Frankfurt/Ma & Boulder, CO: Campus/Westview Press, 173-190.

Franklin, S. & Graesser, A., 1996. "Is it an Agent, or just a Program?: A Taxonomy of Autonomous Agents," Proceedings of the Third International Workshop on Age Theories, Architectures, and languages. Heidelberg: Springer

Guilfoyle, C., Jeffcoate, J. & Stark, H. (eds.), 1997. Agents on the Web. Catalyst for commerce. Ovum Reports, Ovum Ltd., London.

Hennen, L., 1992. Technisierung des Alltags. Ein handlungstheoretischer Beitrag zur Theorie technischer Vergesellschaftung. Opladen: Westdeutscher Verlag.

Hofmann, J., 1994. "Two Versions of the Same: The Text Editor and the Automatic Letter Writer as Contrasting Conceptions of Digital Writing", in: A. Adams, J. Emms, E. Gree & J. Owen (eds.), Women, Work and Computerization. Breaking Old Boundaries - Building New Forms. IFIP, North Holland, 129-142.

Hofmann, J., 1997. "Über Nutzerbilder in Textverarbeitungsprogrammen. Drei Fallbeispiele", in: M. Dierkes (ed.), Technikgenese. Befunde aus einem Forschungsprogramm. Berlin edition sigma, 71-97.

Horberg, J. K., 1996. "Future Agents," in: J. Williams (ed.) Bots and other Internet Beasties. Indianapolis: Sams.net Publishing, 55-74.

Kubicek, H. & Rolf, A., 1986. Mikropolis. Mit Computernetzen in die "Informationsgesellschaft". Hamburg: VSA.

Latour, B., 1996. Aramis, or the Love of Technology. Cambridge, MA & London, UK Harvard University Press (2nd edition).

Latour, B., 1997. The Trouble With Actor-Network Theory. A Free Clarifications Plus More Than a Few Complications. Ms. Paris/San Diego

Law, J., 1997. Traduction/Trahison: Notes On ANT, Centre for Social Theory and Technology, Keele University. http://www.keele. ac.uk/depts/stt/staff/jl/pubs-JL2.htm.

Maes, P. 1994 "Agents that reduce work and information overload", in: Proceedings of the CACM '94

Negroponte, N., 1995. Being Digital. New York: Knopf.

Sargent, 1992"Back to School for a Brand New ABC?", The Guardian, 12 March 1992, 28.

Seybold 1982 "The Wangwriter. Stand-alone that Missed the Mark?", in: The Seybold Report on Office Systems.

Wooldridge, M. & Jennings, N., 1995, "Intelligent Agents: Theory and Practice," Knowledge Engineering Review, Volume 10, No2.

1 To some extent, the social study of technology has made visible the influence that the ideas of software developers as to the use(r)s have on the way software is designed. A study on the subject of word processing programmes (Hofmann 1994; Hofmann 1997) for example showed that from the manufacturer's point of view, the typists' requirements are the primal concern when designing software. Imaginary users not only manifest themselves in programmes but also in the organisation of computer-aided procedures: certain types of text production are facilitated, others are made more difficult or impossible.

2 Actor-network theory (ANT) is a body of theoretical and empirical writing (Actor Network Resource 1997) that grew up in science and technology studies and has become perhaps the most influential approach in this field. Recently, actor network approaches have started to become popular in other domains such as geography, social anthropology and management studies. (Law 1997). At a very general level, ANT may be characterised as a conceptual resource to deal with agencies and entity building. It focuses on entities called actor networks. There is no assumption that specific links or nodes in an actor network are guaranteed. Actor network theorists are concerned with the attribution of human and non-human characteristics across actors; the distribution of properties among those actors; the connections established between them; the circulation entailed by these attributions, distributions and connections, and the transformation of those attributions, distributions and connections (Latour 1997). Thus, the analysis of ordering struggle is central to ANT. An actor network in the sense of ANT may or may not be a technical network.

3 The idea that a technical artifact can be described as a scenario or script has largely been developed by Madeleine Akrich (Akrich 1992a; Akrich 1992b).

4 As an effort in scenarisation that has been influential on the popular imagination concerning intelligent agents (amongst other things) cf. Negroponte 1995.

5 quoted in: Kubicek & Rolf 1986, p. 43.

6 The transformation of textual into technical programs of actions leaves room to a range of substitutions between human and non-human agency and different kinds of materiality. In other words, programs of action can be redistibuted and the directions of this redistributions are many. For example, a systems , designer may shift the competence IS AUTHORIZED TO OPEN THE DOOR either inside a key or inside a memorized code; ... the task of opening the door may be either shifted to humans or to nonhumans (through the figurative attribution of electronic eyes); the basic competence of opening the door may be written down through instructions (linguistic level) as for airplanes, or shifted to the pragmatic level (emergency one-way exit doors that open when pressed upon by a panicked crowd)." (Akrich & Latour 1992, p.262)

7 Although this is still utopian to a certain extent, for example a bathroom-cleaning machine or an automatic window-cleaner; in short, the self-cleaning house, some elements are taken from existing examples such as the housewives' club (http://www.icfde/hausfrau/).

8 8 "Carl Hewitt recently remarked that the question what is an agent? is embarrassing for the agent-based community in just the same way that the question what is intelligence? is embarrassing for the mainstream AI community. The problem is that although the term is widely used by many people working in closely related areas, it defies attempts to produce a single universally accepted definition. This need not necessarily be a problem: after all, if many people are successfully developing interesting and useful applications, then it hardly matters that they do not agree on potentially trivial terminological details." (Wooldridge & Jennings, 1995)

9 For a detailed description of how one family worked with Bob, see http://wwwl.zdnet.com/familypc/familyt/ftsoft/951 I/bob.html


home page about us documents miscellaneous sitemap
home page about us documents miscellaneous sitemap

Projektgruppe "Kulturraum Internet". c/o Wissenschaftszentrum Berlin für Sozialforschung (WZB)
Reichpietschufer 50, 10785 Berlin. Telefon: (030) 254 91 - 207; Fax: (030) 254 91 - 209;
; http://duplox.wzb.eu.